firecrawl-research-patterns

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 1.00). The skill's workflow (SKILL.md and references/api-endpoint-reference.md) explicitly calls Firecrawl's /v1/search and /v1/scrape to fetch and persist scraped markdown from public sources (e.g., arxiv.org, Semantic Scholar, author blogs, IEEE/ACM pages), and those raw third‑party pages are read by the LLM to extract learnings, generate follow-up queries, and drive further scraping—creating a clear vector for indirect prompt injection.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill performs runtime fetches to the self-hosted Firecrawl API (e.g., http://172.25.236.1:3002/v1/search and /v1/scrape, plus the scraper wrapper at http://172.25.236.1:3003) to retrieve raw markdown that is persisted and then injected into LLM prompts, so remote content directly controls model input and the workflow depends on that endpoint.